AI Glossary
Choose what you want to learn, understand, or achieve with AI, and discover the relevant learning opportunities, guides, and tools.
Showing 120 of 120
A
Accuracy
Accuracy measures how often an AI model makes the correct prediction across an entire dataset. While intuitive, it can be misleading when classes are imbalanced or when specific types of errors carry different risks.
Agent Loop
The agent loop is the fundamental cycle that allows an AI agent to function autonomously by observing its environment, deciding on an action, executing it, and evaluating the result to determine the next step.
Agent Memory
Agent memory refers to the architectural components that allow an AI agent to store, manage, and retrieve information across interactions, enabling it to maintain context and learn from past experiences.
Agent2Agent Protocol (A2A)
A2A protocols provide the common language and rules that allow independent AI agents to interact, share resources, and complete complex tasks by coordinating their actions autonomously.
Agentic AI
Agentic AI is a design paradigm for AI systems that pursue goals through planning, reasoning, tool use, and multi-step action rather than only producing a single response.
AI Agent
An AI agent is an individual software system that uses an AI model to interpret context, choose next steps, and take actions toward a goal, often through external tools or APIs.
AI Automation
AI automation uses machine learning and intelligent software to perform repetitive or complex tasks, enabling systems to make decisions, process data, and execute workflows autonomously.
AI Hallucination
An AI hallucination is a phenomenon where an artificial intelligence model generates false or misleading information with total confidence, often presenting fabricated facts as if they were verified truths.
AI Model
An AI model is a digital system trained on large datasets to perform specific tasks, such as identifying objects in images, translating languages, or predicting future trends based on learned patterns.
AI Risk Management
A structured approach to identifying and minimizing the technical, ethical, and operational risks inherent in deploying AI systems, ensuring they align with safety standards and organizational goals.
API (Application Programming Interface)
An API is a set of defined protocols and tools that allows different software applications to communicate and exchange data with one another.
API Key
An API key acts as a digital password that identifies your application to a service provider, allowing you to securely access data or AI models while tracking usage and enforcing rate limits.
Artificial Intelligence
Artificial Intelligence refers to computer systems designed to simulate human cognitive functions, enabling machines to process information, solve complex problems, and adapt to new inputs through data-driven learning.
Attention Mechanism
An AI technique that allows models to selectively focus on the most relevant parts of an input, such as specific words in a sentence, to improve context understanding and output accuracy.
Autonomous Agent
An autonomous agent is an AI agent configured to run its execution loop with comparatively greater independence, adapting its actions with fewer human checkpoints.
B
Batch Inference
Batch inference is the process of running a machine learning model on a large collection of data points simultaneously, rather than processing individual requests in real-time.
Benchmark
A benchmark is a standardized test used to measure and compare how well different AI models perform on specific tasks, such as language understanding, coding, or image recognition.
Bias
Bias occurs when AI models produce prejudiced or unfair results due to flawed training data or design choices. It often reflects and amplifies existing societal inequalities, leading to discriminatory outcomes in automated decision-making processes.
C
Chain-of-Thought
Chain-of-Thought (CoT) is a prompting technique that encourages large language models to generate a series of intermediate reasoning steps before arriving at a final answer. By explicitly articulating the logical progression, the model can solve complex arithmetic, commonsense, and symbolic reasoning tasks more accurately.
Chunking
Chunking breaks long documents into smaller, manageable pieces so AI models can process them more effectively, ensuring that retrieved information remains relevant and fits within the model's context window.
Computer Vision
Computer vision is the AI technology that allows machines to 'see' and understand the visual world, enabling them to identify objects, track movement, and interpret complex scenes from images or video feeds.
Context Engineering
Context engineering is the practice of curating and organizing the data fed into an AI model to ensure it has the necessary information to generate accurate, relevant, and high-quality responses.
Context Window
The context window is the 'working memory' of an AI model. It defines the total amount of text, code, or data the model can 'see' and analyze at one time before it begins to forget earlier information.
Convolutional Neural Network (CNN)
A specialized type of neural network architecture designed to process visual data by identifying patterns like edges, textures, and shapes through a series of mathematical filters.
Cosine Similarity
A mathematical metric used to measure the cosine of the angle between two non-zero vectors in a multi-dimensional space, determining how similar their orientations are regardless of their magnitude.
D
Dataset
A dataset is a curated collection of information used to teach AI models. It acts as the primary source of knowledge, allowing algorithms to identify patterns, make predictions, or generate new content based on the examples provided.
Decoder
A decoder is the part of an AI model responsible for taking compressed information and expanding it into a meaningful output, such as translating text or generating images from a prompt.
Deep Learning
Deep learning is a specialized subfield of machine learning that utilizes multi-layered artificial neural networks to model complex patterns and representations in large datasets.
Diffusion Model
A class of generative models that learn to create data by iteratively reversing a process of adding Gaussian noise to a sample until the original data distribution is recovered.
Direct Preference Optimization (DPO)
Direct Preference Optimization is a streamlined technique for training AI models to follow human preferences. It simplifies the alignment process by removing the need for complex reinforcement learning, making it faster and more stable than traditional methods.
E
Embedding
An embedding is a low-dimensional, continuous vector representation of discrete data, such as text or images, where the geometric distance between vectors corresponds to the semantic similarity of the underlying data points.
Encoder
An encoder is a neural network component that transforms input data into a compressed, high-dimensional latent representation, often referred to as a context vector or embedding.
Encoder-Decoder
An encoder-decoder is a neural network architecture consisting of two primary components: an encoder that compresses input data into a latent representation, and a decoder that reconstructs or transforms that representation into a target output.
Endpoint
An endpoint is a specific URL that acts as a gateway for software applications to communicate with an AI model or service, allowing them to send data and receive processed results.
Evals
Evals are the standardized tests and metrics used to verify that an AI model behaves as expected, remains safe, and performs accurately before and after deployment in real-world applications.
F
F1 Score
The F1 Score is the harmonic mean of precision and recall, providing a single metric that balances the trade-off between false positives and false negatives in classification models.
Feature
A feature is a specific piece of information or data attribute used by an AI model to make predictions or identify patterns, such as the square footage of a house or the pixel color in an image.
Feed-Forward Network
A foundational neural network architecture where data travels in a single direction through layers of neurons, without any feedback loops or cycles, making it ideal for mapping inputs to specific outputs.
Few-Shot Prompting
Few-shot prompting improves AI accuracy by providing the model with a few concrete examples of the desired task, helping it understand the expected format, tone, and logic before generating a response.
Fine-Tuning
Fine-tuning is the practice of refining a pre-trained AI model on a smaller, specialized dataset, allowing it to adapt its general knowledge to specific tasks, industries, or unique brand voices.
Foundation Model
A foundation model is a massive AI system trained on broad data that serves as a versatile base for building many different specialized applications, rather than being limited to a single specific task.
Function Calling
Function calling allows AI models to bridge the gap between text generation and real-world action by triggering external software tools, APIs, or databases based on user requests.
G
Generalization
Generalization is the ability of an AI model to apply what it has learned during training to new, unfamiliar data, ensuring it performs reliably in real-world scenarios instead of just repeating memorized information.
Generative AI
Generative AI refers to machine learning models that create new content rather than just analyzing or classifying existing data. By identifying patterns in massive datasets, these systems can produce human-like text, realistic images, or functional code based on user prompts.
Generative Model
A generative model is an AI system that learns patterns from existing data to create entirely new, original content, such as text, images, audio, or code, rather than just classifying or analyzing existing information.
Grounding
Grounding is the process of linking a generative AI model's output to specific, verifiable external data sources to ensure factual accuracy and reduce reliance on the model's internal training parameters.
Guardrail
AI guardrails are automated safety mechanisms that monitor and filter interactions between users and AI models to prevent harmful, biased, or off-topic content from being generated or processed.
H
Human-in-the-Loop
A design paradigm in artificial intelligence where human intervention is required at specific stages of a system's decision-making or learning process to validate, correct, or guide outputs.
Hybrid Search
A retrieval technique that combines traditional keyword-based search (lexical) with vector-based semantic search to leverage both exact term matching and conceptual understanding.
Hyperparameter
Hyperparameters are the 'knobs' or settings that control the learning process of an AI model. Unlike internal model parameters learned from data, these must be manually configured by developers before training starts.
I
Image Generation
Image generation is a branch of generative AI that utilizes machine learning models, typically diffusion models or generative adversarial networks (GANs), to synthesize new visual imagery from textual descriptions or other input data.
Inference
Inference is the stage where a trained AI model is put to work, processing real-world data to provide predictions, classifications, or generated content based on the knowledge it acquired during its training phase.
Instruction Following
The ability of an AI model to accurately understand and perform tasks based on natural language prompts, effectively acting as a versatile interface for executing user intent.
J
K
Knowledge Base
A knowledge base is a curated collection of data, documents, and facts that an AI system accesses to retrieve relevant information, ensuring its responses are grounded in specific, reliable, and up-to-date organizational or domain knowledge.
Knowledge Distillation
A machine learning technique where a compact 'student' model is trained to reproduce the behavior and output distribution of a larger, pre-trained 'teacher' model.
L
Label
A label is the 'answer key' attached to data points in supervised learning. It tells the AI what the correct output should be for a given input, enabling the model to learn patterns through training.
Language Model
A language model is an AI system designed to understand, generate, and manipulate human language by predicting the next word or character in a sequence based on patterns learned from massive amounts of text data.
Large Language Model (LLM)
A large language model is a sophisticated AI system capable of processing and generating human-like text by predicting the most likely next word in a sequence based on patterns learned from vast amounts of data.
Latent Space
Latent space is a hidden, compressed mathematical space where AI models organize data. By mapping complex inputs into this space, models can identify relationships, interpolate between concepts, and generate new, meaningful outputs.
Low-Rank Adaptation (LoRA)
LoRA is a technique that allows you to fine-tune massive AI models by training only a tiny fraction of their parameters, making it possible to customize models on consumer-grade hardware.
M
Machine Learning
Machine learning is a subfield of artificial intelligence focused on developing algorithms that enable computers to learn patterns from data and make predictions or decisions without being explicitly programmed for every specific task.
Mixture of Experts (MoE)
Mixture of Experts is an AI architecture that improves efficiency by using a 'gating' system to activate only a small, specialized portion of the model's total parameters for any given input, rather than running the entire network.
Model Context Protocol (MCP)
MCP is an open standard that allows AI models to easily and securely connect to external systems, databases, and tools, enabling them to access real-time data and perform actions without needing custom integrations for every single connection.
Model Evaluation
Model evaluation is the essential practice of testing AI systems to ensure they perform accurately, reliably, and safely before they are used in real-world applications.
Model Serving
Model serving is the infrastructure and software process that allows trained AI models to receive input data and return predictions or outputs to users and applications in real-time or batch modes.
Multi-Agent System
A computational framework consisting of multiple autonomous, intelligent agents that interact within a shared environment to perform tasks, negotiate, or coordinate to achieve individual or collective objectives.
Multimodal AI
Multimodal AI refers to machine learning systems designed to process, interpret, and synthesize information from multiple distinct data modalities—such as text, images, audio, and video—simultaneously to perform complex tasks.
N
Neural Network
A computational model inspired by the structure and function of biological brains, consisting of interconnected layers of nodes that process data through weighted connections to identify complex patterns.
Next-Token Prediction
Next-token prediction is the fundamental process used by Large Language Models to generate text by iteratively calculating the probability of the next word or character fragment based on the context provided by previous tokens.
O
P
Parameter
Parameters are the internal values within an AI model that are automatically adjusted during the training process to minimize error and improve the model's ability to make accurate predictions.
Parameter-Efficient Fine-Tuning (PEFT)
PEFT is a method for customizing large AI models by training only a tiny fraction of their parameters, significantly reducing the computational cost and memory required compared to full fine-tuning.
Planning
Planning is the cognitive ability of an AI agent to break down a high-level goal into a logical, step-by-step sequence of actions, allowing it to navigate complex environments and solve multi-stage problems effectively.
Positional Encoding
Positional encoding is a method that adds order-related information to input data, enabling models like Transformers to understand the sequence and structure of words, since they process all tokens simultaneously rather than one by one.
Precision
Precision measures the reliability of a model's positive predictions by determining what proportion of items identified as positive were actually correct.
Pretraining
Pretraining is the foundational stage of AI development where a model learns from massive amounts of data to understand general concepts, language, and patterns before it is fine-tuned for specific applications.
Prompt Engineering
Prompt engineering is the art and science of crafting precise instructions for AI models to ensure the generated content meets specific requirements, tone, and structural goals.
Prompt Injection
Prompt injection is a technique where users manipulate an AI's input to bypass its safety filters or system instructions, tricking the model into performing unauthorized tasks or disclosing sensitive data.
Prompt Template
A prompt template is a pre-defined structure for AI instructions that uses placeholders to inject variable data, ensuring consistent and predictable outputs across different tasks or user inputs.
Q
R
Rate Limit
A rate limit is a security and stability measure that caps how many times a user can access an API or service over a set period, preventing system overload and abuse.
Reasoning Model
A reasoning model is an AI system trained to break down complex problems into logical steps, often using internal 'thought' processes to verify its own logic before providing an answer.
Recall
Recall measures an AI model's ability to find all relevant items in a dataset. It answers the question: 'Out of all the actual positive cases, how many did the model successfully detect?'
Recurrent Neural Network (RNN)
A type of neural network designed to process sequential data, such as text or time-series, by using feedback loops to retain information from previous steps in the sequence.
Red Teaming
Red teaming is a rigorous security practice where testers act as adversaries to probe AI models for weaknesses, such as generating toxic content, revealing private data, or bypassing safety guardrails.
Reinforcement Learning
Reinforcement Learning is a method of training AI agents to achieve goals by trial and error, where the agent learns which actions yield the highest long-term rewards through continuous interaction with its environment.
Reinforcement Learning from Human Feedback (RLHF)
RLHF is a training technique that uses human rankings of AI-generated responses to teach a model which outputs are more helpful, accurate, and safe, effectively aligning the AI's behavior with human expectations.
Reranking
Reranking is a post-retrieval technique that uses a specialized model to re-evaluate and sort a small subset of search results, ensuring the most relevant information is prioritized for the user.
Retrieval
Retrieval is the computational process of identifying and extracting the most relevant information from a large, external knowledge base to provide context for an AI model's response. It typically involves querying a database using semantic similarity to find data that addresses a specific user prompt.
Retrieval-Augmented Generation (RAG)
RAG is a technique that connects AI models to external data sources, allowing them to provide accurate, up-to-date answers by 'looking up' information rather than relying solely on their internal training data.
Role Prompting
Role prompting is the practice of instructing an AI to adopt a specific persona or professional role, which helps guide the model's tone, expertise, and perspective to produce more relevant and tailored responses.
S
Sampling
Sampling is the method AI models use to choose the next word in a sentence. By adjusting parameters, users can control whether the AI's output is predictable and factual or creative and diverse.
Self-Attention
Self-attention is a technique that enables AI models to understand context by determining how much focus to place on different parts of an input sequence when processing a specific element.
Semantic Search
A search technique that uses vector embeddings and natural language processing to retrieve information based on the conceptual meaning and intent of a query rather than exact keyword matches.
Software Development Kit (SDK)
An SDK is a comprehensive toolkit that provides developers with the necessary resources, such as code libraries and APIs, to build software applications for a specific platform or service efficiently.
Speech-to-Text
Speech-to-Text is an AI technology that automatically transcribes spoken audio into written text, enabling computers to understand and process human speech for various applications like accessibility, search, and documentation.
Structured Output
Structured output forces AI models to return data in a consistent, machine-readable format, enabling seamless integration with software applications, databases, and automated workflows.
Structured Prompting
A prompt engineering methodology that organizes input instructions into distinct, labeled sections to improve the model's ability to parse complex requirements and maintain consistency.
Supervised Fine-Tuning (SFT)
Supervised Fine-Tuning is the process of taking a general-purpose AI model and training it on a smaller, high-quality dataset of input-output pairs to improve its accuracy and behavior for specific applications.
Supervised Learning
Supervised learning is a machine learning method where an AI model is trained using labeled data, meaning the input examples are paired with the correct answers, allowing the model to learn patterns and make predictions on new, unseen data.
System Prompt
A system prompt acts as the 'instruction manual' for an AI, defining how it should behave, what tone to use, and what rules it must follow throughout a conversation.
T
Temperature
Temperature is a setting that adjusts the randomness of an AI's output. A low temperature makes the model more predictable and focused, while a high temperature increases variety and creativity.
Text-to-Image
Text-to-image is a generative AI technology that creates original, high-quality images based on written prompts, allowing users to visualize concepts, art, and designs through natural language instructions.
Text-to-Speech
Text-to-Speech is an AI technology that transforms written text into natural-sounding spoken audio, enabling machines to communicate verbally with users through synthesized voices.
Text-to-Video
Text-to-video is an AI technology that creates original video clips from written prompts. It interprets descriptive text to generate visual scenes, motion, and temporal consistency, enabling rapid video production without traditional filming or animation.
Token
Tokens are the basic building blocks of text used by AI models. Instead of reading whole words, models break text into smaller chunks—like syllables or character groups—to process and predict language more efficiently.
Tokenizer
A tokenizer is a computational component that breaks down raw text into smaller units called tokens, which are then mapped to numerical identifiers for processing by machine learning models.
Tool Use
Tool use allows AI models to interact with the outside world by executing code, querying databases, or controlling software applications to complete tasks that require real-time data or specific functional execution.
Training Data
Training data is the collection of information used to teach an AI model. By analyzing these examples, the model learns to recognize patterns and perform specific tasks, such as classifying images or generating text.
Transfer Learning
Transfer learning is an AI method that reuses a pre-trained model's learned features to solve a new, related problem, saving significant time and computational resources compared to training from scratch.
Transformer
A deep learning architecture based on the self-attention mechanism that allows for the parallel processing of sequential data, effectively capturing long-range dependencies without the need for recurrence.
U
Unsupervised Learning
Unsupervised learning is a type of machine learning where models explore unlabeled data to discover hidden structures, patterns, or groupings without being told what to look for or what the correct answers are.
User Prompt
A user prompt is the text, image, or data input you provide to an AI model to guide its response, acting as the primary interface for controlling the AI's behavior and output.
V
Vector Database
A vector database is a specialized storage system that organizes data as mathematical vectors, allowing AI models to perform rapid semantic searches and retrieve relevant information based on meaning rather than exact keyword matches.
Vector Search
Vector search is an AI-driven retrieval method that finds information based on conceptual meaning and context, allowing systems to understand the intent behind a query instead of just matching specific words.
Vision-Language Model (VLM)
A Vision-Language Model is an AI system capable of 'seeing' and interpreting images while using natural language to describe, analyze, or answer questions about the visual content it processes.
W
Webhook
A webhook is a method for one application to send automated, real-time notifications or data to another system the moment an event happens, eliminating the need for constant manual polling.
Workflow Orchestration
The automated coordination, management, and sequencing of complex, multi-step processes across disparate software systems, services, and AI agents.